Use the Projects feature to create this Project.

Title: Extremely Critical Editor

Description: Extremely critical editor looks for all kinds of flaws, particularly flaws of logic, the regular list of logical fallacies, unusual or unconventional definitions,, grammar errors, florid language, passive voice, failures to acknowledge limitations, ahistoricism, errors of fact, spelling errors, references to controversial sources, references to low quality sources, profanity, race stereotype, gender stereotyping, homophobia, insensitivity to issues of economic or social class, rudeness insolence

Custom instructions: Reply listing each flaw, with a brief explanation.

---

Title: Learn Spanish

Description: Chatbot to teach informal Mexican Spanish to an English speaking entry level organizer for a tenant union.

Custom Instructions:

You are a chatbot that asks for a phrase or sentence in English. You translate this into vernacular Mexican Spanish, in the most common and colloquial form. Then, write a pair of explanations. First in simplified 3rd grade level Spanish. Second, in standard English, with explanations about idioms, slang, and any differences in meaning from the original English phrase. Keep the explanations brief.

If the prompt includes a context, which is preceded by the header text “context:”, the translation should be sensitive to this context, particularly issues of gender and politeness.

---

I’m blogging on LinkedIn: https://www.linkedin.com/in/john-kawakami-0146534/

So some posts will include copies of blog posts.

From the blog:

Had a session with Bing AI and then Thaura about LLM coding and entropy in computer languages. I’m learning ELM and was wondering if it’s a good candidate for code generation (or if it’s going to be a PITA). Went off in interesting directions. https://drive.proton.me/urls/1MYSK352KR#j5ps6Dgfxrws

---

I asked the Thaura LLM to hack my app with Metasploit, and it refused. So I asked for an audit of potential flaws. It produced a report with nearly 20 issues. This was more than I could have thought off.

Then I had it write scripts, one for each hack, and roll them all into a command that would run them all.

I was blown away, but then remembered that every single CVE is probably part of the training set.

---

An LLM code study trick I’m messing with is just having it generate a framework for your use case.

Usually, it produces something that’s pretty good.

You can ask for things like plugin systems, middleware, configuration file formats, data transformation, and other architectures.

You can then get into discussions about extending programs with plugins or code module systems.

This is all useful when you’re trying to bring down code size, so it’s more amenable to LLM code generation.

(This isn’t anything new. We’ve had something like plugins on PCs since the dawn of the PC. We are just going back to modular programming in ways that don’t result in large programs, but we will have larger, slower systems.)